# X's 2026 algorithm update: why shares now outweigh likes > X open-sourced its ranking weights in 2026: a copy-link share scores 20 vs 0.5 for a like. Here's what that 40x gap actually means for what you post. - Published: 2026-09-25 - Last updated: 2026-09-25 - Author: Sanju (Founder & CEO @ DunSocial) - Topics: x, strategy --- X's own open-sourced ranking code says a copy-link share is worth 40 times what a like is worth in its For You feed. That's not a rumor or a growth-hacker's guess. It's a number in a file X put on GitHub itself. If your posting habits are still built around chasing likes, you're optimizing for the wrong metric on X in 2026. ## Short answer On **13 August 2026**, X published the actual weights its For You ranking model uses to score predicted actions, in `home-mixer/params/param.rs` inside its open-source [`x-algorithm`](https://github.com/xai-org/x-algorithm) repository. A **copy-link share scores 20.0** & a **like scores 0.5**, a 40x gap, while a **report scores -234.0**. X has kept shipping updates to that repository through September, including a bigger boost for accounts under 1,000 followers & a new dwell-time signal. X's own README adds an important caveat: these are multipliers on your *predicted probability* of taking an action, not literal multipliers on raw engagement counts. So "1 report cancels out 468 likes" is a popular but technically wrong reading of the numbers. Practically: write things people want to pass on, not just tap a heart on. A post worth copying & sending to a friend or a group chat is worth more to your reach than ten posts that only collect likes. ## What X actually published, & when | Predicted action | Weight | What it means | | --- | --- | --- | | Share via copy link | 20.0 | The single largest positive weight in the file | | Reply | 5.0 | Meaningfully rewarded | | Quote | 5.0 | Same tier as a reply | | Share via DM | 5.0 | Sending a post privately counts as much as replying publicly | | Follow the author | 4.0 | A post good enough to convert a reader into a follower | | Repost | 1.0 | Worth twice a like, far less than a link share | | Like | 0.5 | The smallest positive engagement weight tracked | | Report | -234.0 | The largest negative weight in the file | | Block author | -31.2 | A strong negative signal | | Mute author | -58.8 | Even stronger than a block | Source: [ppc.land's analysis](https://ppc.land/x-gives-copy-link-shares-a-feed-weight-of-20-against-0-5-for-likes/) of the file X published, cross-checked against the [`xai-org/x-algorithm`](https://github.com/xai-org/x-algorithm) repository's own README. X has continued to update the ranker weekly since the August release. A September 16 change added a separate reply-spam model for large accounts & a 10-second click-dwell signal, per [keywordseverywhere.com's ongoing tracker](https://keywordseverywhere.com/news/social-algorithm-updates/). **The caveat X wants you to get right:** its own README states these weights "scale the predicted probabilities of such actions... they do *not* scale the raw engagement counts." It also calls reading a report as "468 times worse than a like" a common misconception the company added code comments to correct. Treat the table above as *which actions the ranker rewards most*, not as an exact exchange rate. ## Why the timing matters This isn't an isolated ranking tweak. On **7 September 2026**, X shut down its old Creator Revenue Sharing program & replaced it with an Original Content Rewards program that explicitly excludes reposts, lightly-edited reshares, & engagement-bait. It rewards only content X can verify a creator actually made. Read together with the ranking weights, the direction is consistent: X is pushing both its algorithm & its monetization rules toward original posts people choose to pass along, & away from volume-based reply-bait & recycled content. ## What to actually do about it - **Write the kind of post someone forwards, not just likes.** A useful explainer, a strong opinion, or a genuinely funny observation earns a copy-link share. A post that only works as a quick reaction earns a like & stops there. - **Don't mistake volume for reach.** X's out-of-network discount & repeated-author decay mean spamming the feed with many similar posts a day gives each one a shrinking chance of spreading. Quality beats cadence. - **Threads still work, if they hold together.** A thread people read end to end & share as a whole benefits from the same share-weighted ranking as a single post. DunSocial composes it as one connected chain, not a pile of orphaned replies. - **Don't chase reply-guy tactics on big accounts.** Reply-spam detection on large accounts got its own model in the September 16 release, so a spammy reply aimed at a big account's audience is now more likely to get flagged than seen. - **Keep your voice consistent.** A post still has to sound like you. Saved brand facts & a fixed Voice Preset help a team stay recognizable across a lot of individually shareable posts instead of drifting per draft. See [stop AI posts sounding generic](/blog/stop-ai-posts-sounding-generic). ## How DunSocial helps you write for shares Writing a post worth sharing takes more thought than a throwaway reaction. DunSocial makes that the fast part: AI Memory & your Voice Preset are pulled into every X draft, so a considered, on-brand post doesn't take longer to write than a quick one. Full X threads compose as a single connected chain of up to 25 posts, & a human confirms every post before it goes out. See [best social media scheduler for X](/blog/best-social-media-scheduler-for-x). The best-time-to-post heatmap, built from your own account's activity, shows when your audience is around so your best posts land in front of the people most likely to share them. $20/mo or $192/yr, 14-day trial, no card. [Start scheduling to X](https://app.dunsocial.com/login). ## FAQ ### Is "1 report equals 468 likes" true? Not the way it's usually stated. That ratio comes from dividing the report weight (-234.0) by the like weight (0.5), but X's own README says the weights scale *predicted probabilities* of each action, not raw counts. So it isn't a literal exchange rate. Read it as "reports hurt a post's reach far more than likes help it," not as exact arithmetic. ### Does this replace posting consistently? No. X's out-of-network discount & repeated-author decay reward a healthy rhythm, but reward each individual post less if you flood the feed with many similar ones in a short window. The change is about what kind of post to write, not whether to keep a schedule at all. ### Should I stop trying to get replies? No. A reply is still weighted 10x a like (5.0 vs 0.5). The change to watch is that large-account reply-spam now gets its own detection model. A reply written to add something is fine; a reply written purely to farm visibility on someone else's audience is more likely to get flagged. ### Where can I read the actual numbers myself? Directly in X's own open-source repository: [`github.com/xai-org/x-algorithm`](https://github.com/xai-org/x-algorithm), specifically `home-mixer/params/param.rs`. It's public, Apache-2.0 licensed, & updated regularly. ## Sources - [xai-org/x-algorithm](https://github.com/xai-org/x-algorithm): X's own open-source For You ranking repository - [ppc.land: "X gives copy-link shares a feed weight of 20, against 0.5 for likes"](https://ppc.land/x-gives-copy-link-shares-a-feed-weight-of-20-against-0-5-for-likes/) (21 September 2026) - [keywordseverywhere.com: "Social Media Algorithm Updates 2026"](https://keywordseverywhere.com/news/social-algorithm-updates/): ongoing tracker of X, Reddit, TikTok, Pinterest, Facebook & LinkedIn ranking changes - [postforsuccess.com: "X Creator Revenue Sharing Ends Sept 7"](https://postforsuccess.com/x-creator-revenue-sharing-ends-2026) Retrieved 25 September 2026. --- ## About this document This is the Markdown representation of https://www.dunsocial.com/blog/x-algorithm-update-2026-shares-outweigh-likes. The HTML version of the same page is at the same URL. You can also request Markdown from any page by sending `Accept: text/markdown`. Machine-readable summary of the whole site: https://www.dunsocial.com/llms.txt ## Company - Product: DunSocial, https://www.dunsocial.com - Legal entity: THISUX PRIVATE LIMITED, Chennai, Tamil Nadu, India - Support: support@dunsocial.com - Open the app: https://app.dunsocial.com/login